Acoustic emissions are ultrasonic waves emitted by a material when it undergoes irreversible damage, which can be used to non-destructively perform life-cycle analysis on safety-critical components. Machine learning (ML) has been used as a means of classifying the damage mechanism of the source of acoustic emissions for decades but acquiring enough data to train ML models is challenging. Pencil lead was broken in three orientations on an aluminium plate to simulate acoustic emissions in order to generate a dataset that will be used to train machine learning models, which will classify waveforms based on the orientation they were broken in. In order for ML algorithms to be used in the field in the future, they need to be robust to a wide variety of experimental variables, such as boundary conditions, location of the event, and the interface between the sensor and the material. By training ML algorithms on a carefully selected subset of scenarios, high accuracy is able to be maintained while keeping data requirements relatively low. The model was able to classify 97.5% of waves correctly, including some taken from experimental conditions it had not seen before.